The goal of osteoporosis treatment is fracture risk reduction. Identification of patients at high risk for fracture is facilitated by the World Health Organization (WHO) diagnostic classification system, whereby osteoporosis is diagnosed in the presence of a T-score of −2.5 or less. However, many fragility fractures occur in individuals with a bone mineral density (BMD) T-score that is better than −2.5.1 As such, limiting treatment to patients with a T-score of −2.5 or less would miss a large number of patients who will later sustain a fracture that might have been prevented by appropriate identification and early intervention. BMD combined with clinical risk factors for fracture provides a better prediction of fracture risk than BMD or clinical risk factors alone. The WHO fracture risk assessment tool, FRAX, is a computer-based algorithm with an input of patient demographics, a “yes” or “no” response indicating the presence or absence of each of seven clinical risk factors for fracture, and femoral neck BMD (when available) to estimate the 10-year probability of major osteoporotic fracture (ie, hip, spine, proximal humerus, and distal forearm) and the 10-year probability of hip fracture.2 Country-specific mortality data can be used to calibrate the fracture probabilities, with 32 countries included in FRAX at the time of this writing. Cost-utility analysis then may be applied to FRAX-derived data, using numerous economic, social, and political assumptions, to determine the magnitude of fracture risk at which it is likely to be cost-effective to initiate pharmacologic therapy to reduce fracture risk. FRAX has been incorporated recently into bone densitometer software and some handheld computer devices, allowing greater access to this tool as an aid in making treatment decisions.
The expression of fracture risk as a probability (absolute fracture risk) provides greater clinical utility than relative fracture risk3 but cannot predict which individual patients will sustain a fracture.4 This is so because fractures are stochastic events (ie, subject to randomness), depending on factors that include falling, the force of the fall, body position when falling, and the nature of the surface that is impacted. Clinicians, patients, and third-party payers must appreciate that the FRAX output is an estimate of fracture probability, not a certainty that a fracture will or will not occur; this caveat is stated directly on the FRAX Web site (www.shef.ac.uk/FRAX/index.jsp). Despite the vast amount of data analyzed to develop the FRAX algorithm and subsequent validation of this tool in approximately 1 million people, it is not surprising that periodic updates to FRAX have been released. It is expected that FRAX will continue to evolve in the future as new data become available.
To this end, the work of Sornay-Rendu and colleagues5 in this issue of JBMR compares FRAX-predicted fracture probability with radiographically confirmed fragility fracture incidence over 10 years in the Os des Femmes de Lyon (OFELY) cohort of 837 French women over the age of 40 years. As expected, in this report, the observed fracture incidence was greater among women with higher FRAX-predicted fracture probabilities. However, fracture prediction was not perfect in that the highest decile of fracture probability identified only 43% of the women who sustained an incident fracture. Moreover, in 50% of the women who sustained a fragility fracture, their estimated fracture probability was less than 10%, which is a value in the “low risk” range according to some guidelines (eg, Osteoporosis Canada, www.osteoporosis.ca) and substantially below the treatment threshold advocated by the US National Osteoporosis Foundation. Thus, approximately half the women who will sustain a fragility fracture are not identified as being at high fracture risk, making it unlikely that they will receive therapy to reduce this risk. Approaches to enhancing the ability of fracture risk tools are desirable.
One important clinical consideration that is not currently incorporated into FRAX is a history of falling. Falls are common with advancing age and may lead to fractures. However, they have not been included in the FRAX calculation because falls history was not collected in all cohorts used to develop the model. Furthermore, falls history was collected differently in various studies such that it was not possible to develop a standardized approach. Clearly, a standardized approach to collecting falls data is needed in cohort studies, perhaps with something as simple as asking, “How many times have you fallen in the least year.” Moreover, a FRAX frequently asked question (FAQ) states, “It is important that risk assessment models identify a risk that can be reduced by treatment” and “although plausible, pharmaceutical intervention has not been shown to reduce fracture risk in patients selected on the basis of a fall history.” Clearly, heterogeneity and the absence of data are valid arguments for noninclusion of falls in fracture risk estimation models. However, the exclusion of risk factors that are not affected by pharmacologic therapy seems to be inconsistent with the inclusion of risk factors such as age and sex, which are also not affected by pharmacologic therapy. In fact, other standardized approaches to fracture risk prediction, for example, the Garvan calculator and the QFractureScore approach (models based on Australian and UK data, respectively), do include falls history.6, 7 Some work suggests that these approaches might improve fracture prediction capability, particularly in older individuals.8–10 Additional refinements of FRAX and other fracture prediction models doubtlessly will occur; these potentially could include other fracture risk factors such as markers of bone turnover and proximal femur geometric parameters.
Guidelines for facilitating the clinical use of fracture risk models are forthcoming. To this end, the International Osteoporosis Foundation (IOF) and the International Society for Clinical Densitometry (ISCD) will conduct the “FRAX Initiative” in late 2010, the goal of which is to facilitate the interpretation and use of FRAX in clinical practice. Moreover, such efforts may help to harmonize some of the fracture risk estimation tools available to clinicians. As an example of differences in fracture risk estimation currently available to clinicians, it is worthwhile to consider a 70-year-old, 130-pound woman from Australia with a femoral neck T-score of −1.8, a history of a spine and wrist fracture after age 50, and one fall within the last year. The estimated 10-year hip fracture risk is 2.2% when using the FRAX calculator compared with 21.2% with the Garvan calculator. Given such a disparity, it seems probable that differing therapeutic recommendations would ensue based on which fracture risk calculator was used. The substantial difference in estimated risk may reflect, at least in part, the consideration of clinical risk factors currently not included in the FRAX model, notably falls and multiple fractures. This patient scenario can be used to explore the importance of these risk factors in the assessment of fracture risk in a given patient. In the preceding example, if one fracture (analogous to the “yes/no” approach used in FRAX) and no falls are entered into the Garvan calculator, the 10-year estimated hip fracture risk is 7.1%; two fractures increase this risk estimation to 15.3%. In the setting of two prior fractures, the inclusion of a single fall leads to the 21.2% hip fracture risk noted earlier; two falls yields a 29% risk, with three or more falls leading to a 38.9% risk. Clearly, clinicians and patients must recognize that clinical tools are imperfect and that clinical practice guidelines may not be applicable for some patients seen in real-world medical care; clinical judgment and individualization of care remain essential.
Another important observation of the study by Sornay-Rendu and colleagues is the finding that 37% (56 of 151) of documented fragility fractures did not occur at the sites of “major” osteoporosis-related fracture (ie, clinical spine, forearm, hip, or shoulder) that are predicted by FRAX. Thus, the observed incidence of osteoporosis-related fractures is substantially higher than that estimated by FRAX. Other causes for FRAX to under- or overestimate fracture risk include the dichotomous (yes/no) input for clinical risk factors that in reality are associated with a range of risk according to dose and duration of exposure (eg, glucocorticoid therapy) and the discordance in BMD compared with the femoral neck (eg, BMD much lower at the lumbar spine than at the femoral neck).
Also in this issue of JBMR, Langsetmo and colleagues11 address the issue of reference database selection for T-scores in men. It is controversial whether a male or female reference database should be used for men, with reasonably good but not overwhelming arguments for both positions. Using data from the Canadian Multicentre Osteoporosis Study (CaMos), Langsetmo and colleagues analyzed three methods of BMD “normalization” in men and concluded that a common (non-sex-adjusted) reference database for T-scores is appropriate for both men and women. This is consistent with the WHO position of using a young white female reference database for T-score calculation for both men and women but at odds with the current ISCD position that recommends a female reference database for women and a male reference database for men. This is perhaps a situation where there is no single correct choice such that it may be better to be consistent than to be “right.” We therefore suggest that consensus or agreement on this issue be achieved through an appropriate venue, with the goal of identifying a single method for calculating T-scores worldwide. T-scores are important for diagnostic classification, estimation of fracture risk, and communication of disease-state information to patients and clinicians; in some countries, such as the United States, diagnostic classification is linked to numerical codes for insurance billing and may play a role in insurance coverage for drug therapy and follow-up BMD testing.
The bottom line is that there is utility in continuing to use T-scores, ideally with a unified method of calculation worldwide. There is also great clinical value in using FRAX to estimate fracture probability, with the understanding that FRAX will continue to evolve as new data become available. Physicians must be fully aware of the benefits and limitations of these clinical tools in order to use them wisely in the care of individual patients.